Linear sufficiency and some applications in multilinear estimation (Q1066565)

In the linear model \(Y=X\beta +u\) the question arises when a linear transformation contains all information in the model. A linear inhomogeneous transformation \(c+Ay\) is called linearly sufficient if there exists a BLUE (best linear unbiased estimator) of Ey which is a linear function of \(c+Ay.\) The paper presents necessary and sufficient conditions for linear sufficiency and linear minimal sufficiency. Then it deals with the computation of expectation and covariance of expressions like \((\epsilon 'A\epsilon)^ 2\), and p-fold Kronecker products. A final section contains results about linear sufficient statistics in multilinear estimation.